Material management intelligent warehouse personnel operation safety management and detection method and system

Abstract

The invention belongs to the technical field of communication installation and maintenance safety management, and particularly discloses a material management intelligent warehouse personnel operation safety management and detection method and system, and the method comprises the following steps: registering the identity information of installation and maintenance personnel and a corresponding operation task, collecting and preprocessing a human body image of the installation and maintenance personnel, and employing a target detection algorithm to detect the operation task of the installation and maintenance personnel. Constructing a target detection model, extracting human body image features, inputting the human body image features into a classification decision tree model, judging whether the extracted features meet specified dressing standards or not, and if the extracted features meet the specified dressing standards, simulating a strong current environment through a preset test device, and detecting alarm functions of the safety helmet and the test pencil; s5, if all safety specifications and tool functions are detected to be qualified, operation is allowed; if the detection result is not qualified, a warning signal is sent out, and the operator is forbidden to continue to work. By adopting the technical scheme, safety standard detection is automatically completed, and installation and maintenance are completed under the condition that safety measures are in place.

Description

Technical Field

[0001] The present invention belongs to the technical field of communication installation and maintenance safety management, and relates to a method and system for intelligent warehouse personnel operation safety management and detection of material management. Background Art

[0002] In the communication installation and maintenance scenario, when front-line personnel go out for work, they need to complete the detection of various safety protection measures. At present, most operations still rely on manual detection and management, resulting in problems such as difficult implementation and supervision of safety standards, leading to inadequate safety protection for operations and frequent occurrence of safety accidents.

[0003] In the prior art, before the communication operator's installation and maintenance personnel go out for work, the following technical solutions are mainly adopted for their operation safety management:

[0004] Manual inspection: Before the installation and maintenance personnel go out for work, the safety management personnel visually inspect the wearing situation of their safety protection devices, record the inspection results using a paper inspection form, and point out and require rectification of the problems found during the inspection on the spot.

[0005] Regular training: Organize the installation and maintenance personnel to participate in regular training on safety protection knowledge and skills, and improve the safety awareness and safety protection ability of the operators through theoretical learning and practical exercises.

[0006] Violation punishment: Formulate safety norms and violation punishment regulations for going out for work. For installation and maintenance personnel who violate safety protection requirements, the safety management personnel shall give warnings, fines, work suspension for rectification and other punishments, and record the punishment results on file.

[0007] However, the prior art solutions have the following deficiencies in practical applications:

[0008] Manual inspection is inefficient, difficult to ensure accuracy, and cannot cover the whole process of going out for work. For the situation of privately removing the protection device after inspection, there is a lack of effective supervision means.

[0009] Regular training is out of touch with the actual operation situation. Some personnel attach importance to learning and neglect practice, and the training effect cannot be fully implemented in daily operations.

[0010] The violation punishment mechanism lags behind the occurrence of problems, lacks necessary preventive and control measures for the existing violations, and the punishment results fail to form a normalized behavior restraint and positive guidance.

[0011] The safety management process is separated from the daily operation process of the installation and maintenance personnel, lacking effective integration, and it is difficult to achieve effective safety supervision and risk prevention and control.

[0012] Insufficient collection and analysis of safety management data of operators, the data value fails to be fully exploited, and it is difficult to achieve digital and intelligent management. Summary of the Invention

[0013] The object of the present invention is to provide a method and system for safety management and detection of personnel operations in an intelligent warehouse for material management, which replaces manual safety assessment and realizes automatic completion of safety standard detection.

[0014] To achieve the above object, the basic solution of the present invention is: A method for safety management and detection of personnel operations in an intelligent warehouse for material management, comprising the following steps:

[0015] S1, registering the identity information of the installation and maintenance personnel and the corresponding operation tasks, collecting the human body images of the installation and maintenance personnel, and performing preprocessing;

[0016] S2, using an object detection algorithm to construct an object detection model, and inputting the preprocessed human body image into the object detection model for recognition and feature extraction;

[0017] S3, inputting the extracted features into a classification decision tree model to determine whether the extracted features meet the specified dressing standards. If the extracted features meet the specified dressing standards, then proceed to step S4; otherwise, output a warning signal to prohibit the continuation of the operation task;

[0018] S4, simulating a strong electric environment through a preset test device to detect the warning functions of safety helmets and electro-test pens;

[0019] S5, if all safety specifications and tool functions are detected to be qualified, then allow the operation; if the detection result is unqualified, then issue a warning signal and prohibit the operator from continuing the operation.

[0020] The working principle and beneficial effects of this basic solution are as follows: This technical solution completes a number of set safety guarantee measures according to the process, and at the same time judges whether they are implemented in place. If the implementation is not standard, the operation cannot continue, ensuring that the installation and maintenance tasks are completed under the condition that the safety measures are in place.

[0021] For the fixed-scene detection of front-line personnel before departure, high-precision judgment is achieved through the dual methods of decision tree + object detection.

[0022] Furthermore, the method of using an object detection algorithm to construct an object detection model and inputting the preprocessed human body image into the object detection model for recognition and feature extraction is as follows:

[0023] Construct multiple object detection models according to requirements to extract features in the image. Each object detection model outputs the position, category information, and confidence level of the corresponding features, obtaining multi-dimensional feature information;

[0024] The features extracted by the object detection model include the wearing situation of safety helmets, the damage situation of safety helmets, the opening and closing situation of work clothes zippers, and the wearing situation of insulating gloves.

[0025] Automatically and regularly monitor and analyze the safety protection situation of maintenance personnel before going out for work, timely discover and warn of potential safety hazards, and reduce the blind spots and loopholes in manual inspections. The clothing specification feature extraction module can be configured with various detection requirements, capable of detecting essential safety helmets and also expanding elements such as work uniform buttons, goggles, insulating gloves, etc.

[0026] Furthermore, the method of inputting the extracted features into the classification decision tree model to determine whether the extracted features meet the specified clothing standards is as follows:

[0027] There is a threshold configuration library in the classification decision tree model, and the threshold or determination rule corresponding to each clothing specification is stored in the threshold configuration library;

[0028] Make a comprehensive judgment on different features.

[0029] Combining the object detection algorithm and the classification decision tree model can make a comprehensive judgment on various clothing and tool features, and the high-precision detection greatly reduces the probability of missed detection and false detection.

[0030] Furthermore, simulate a strong electricity environment through a preset test device to detect the warning functions of safety helmets and electrical test pens. The specific steps are as follows:

[0031] Set a certain intensity of electromagnetic field or near-electricity environment, place the safety helmet and electrical test pen in this environment, and the warning modules on the safety helmet and electrical test pen emit warning sound signals and / or light signals;

[0032] Collect the warning signals emitted by the warning modules on the safety helmet and electrical test pen, and extract the warning signal features;

[0033] Compare the extracted warning signal features with the preset feature range. If the warning signal features are not within this preset feature range, output the abnormal detection information of the safety helmet and electrical test pen, otherwise output the normal detection information of the safety helmet and electrical test pen.

[0034] Through the simulated strong electricity detection method, timely discover faults such as near-electricity warnings or electrical test pen failures, and eliminate potential risks from the source.

[0035] Furthermore, when the target confidence level is lower than the set value or the manual determination of the detection result is incorrect, save the corresponding image or video segment as a difficult example;

[0036] Developers make correct annotations for the collected difficult examples in the background and determine whether to include them in the training data;

[0037] After merging the difficult example library with the original data set, use the incremental training strategy to train and optimize the target detection model or decision tree determination model;

[0038] Replace the old version model with the updated model.

[0039] Perform difficult example automatic collection and model incremental training to ensure that the system can continuously optimize the recognition effect.

[0040] The present invention also provides a safety management and detection system for personnel operations in an intelligent warehouse for material management based on the method described in the present invention, including a dressing code detection module and a safety tool testing module. The dressing code detection module is used to detect whether the dressing of the installation and maintenance personnel is standard, and the safety tool testing module is used to detect whether the functions of the tools carried by the installation and maintenance personnel are normal.

[0041] This system integrates safety detection with the daily operation process of installation and maintenance personnel, automatically checks their safety protection equipment before they go out for work, strengthens the safety management link in the operation process, and ensures that installation and maintenance personnel develop good safety protection habits.

[0042] Further, the dressing code detection module includes a dressing code feature extraction module and a dressing code determination module connected in sequence;

[0043] The dressing code feature extraction module includes a camera, an image preprocessing unit, a target detection model, and a feature data output interface connected in sequence. The camera is used to collect the human body image of the installation and maintenance personnel and transmit it to the image preprocessing unit. The image preprocessing unit is used to perform noise reduction, cropping, and distortion correction on the collected human body image. The target detection model extracts the features of the preprocessed human body image and then conveys them to the dressing code determination module through the feature data output interface;

[0044] The dressing code determination module includes a classification decision tree model, a threshold configuration library, and a determination result output interface. The classification decision tree model is used to determine whether the extracted features meet the dressing standards specified in the threshold configuration library and output the determination result through the determination result output interface.

[0045] After obtaining the static image of the installation and maintenance personnel through the camera, use high-precision target detection algorithms (such as YOLO, R-CNN, etc.) to extract features such as wearing a safety helmet and zipper status for easy use.

[0046] Further, the dressing code detection module further includes a dressing code feature optimization & training module, and the dressing code feature optimization & training module includes a difficult example collection unit, a difficult example screening unit, a model training unit, and a model update unit;

[0047] When the difficult example collection unit detects that the target confidence is lower than the set value or the manual determination of the detection result is incorrect, it saves the corresponding image or video clip as a difficult example;

[0048] The difficult case screening unit makes correct annotations on the collected difficult cases by developers in the background and determines whether to include them in the training data;

[0049] After the model training unit merges the difficult case library and the original data set, it uses the incremental training strategy to train and optimize the object detection model or the decision tree judgment model;

[0050] The model update unit replaces the old version model with the updated model.

[0051] Automatically or manually screen and annotate difficult cases, include them in the incremental training set, and iteratively update the object detection model or the decision tree model to continuously improve the accuracy of clothing detection.

[0052] Furthermore, the safety tool testing module includes a simulated high-voltage device, a microphone acquisition unit, a camera acquisition unit, a data processing unit, and an alarm judgment unit;

[0053] The simulated high-voltage device is provided with a test point output interface. The microphone acquisition unit and the camera acquisition unit are used to collect the alarm signals emitted by the alarm modules on the safety helmet and the voltage detector, and input them into the data processing unit;

[0054] The data processing unit is used to extract the alarm signal features and input them into the alarm judgment unit;

[0055] The alarm judgment unit compares the extracted alarm signal features with the preset feature range to judge whether the functions of the safety helmet and the voltage detector are normal.

[0056] Generate a specific electric field or electromagnetic environment through the simulated high-voltage device to trigger the proximity electric alarm function of the safety helmet. The test process can be completed within a few seconds. Compared with the traditional manual inspection, it is faster and has the same standard, greatly improving the efficiency of pre-operation detection. Brief Description of the Drawings

[0057] Figure 1 It is a schematic flow chart of the intelligent warehouse personnel operation safety management and detection method of the present invention for material management. Detailed Embodiment

[0058] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions from beginning to end. The embodiments described below with reference to the drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation of the present invention.

[0059] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "longitudinal", "transverse", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation to the present invention.

[0060] In the description of the present invention, unless otherwise specified and defined, it should be noted that the terms "installation", "connection", "coupling" should be understood in a broad sense. For example, it can be a mechanical connection or an electrical connection, or it can be the communication inside two elements. It can be directly connected or indirectly connected through an intermediate medium. For those of ordinary skill in the art, the specific meanings of the above terms can be understood according to specific situations.

[0061] The present invention discloses a method for safety management and detection of personnel operations in an intelligent warehouse for material management, which replaces manual safety assessment, can automatically complete safety standard detection, complete a number of set safety guarantee measures according to the process, and at the same time the system judges whether the execution is in place. If the execution is not standard, the operation cannot continue, ensuring that the installation and maintenance tasks are completed under the condition that the safety measures are in place. As Figure 1 shown, the method for safety management and detection of personnel operations in an intelligent warehouse for material management includes the following steps:

[0062] S1, registering the identity information of the installation and maintenance personnel and the corresponding operation tasks, collecting the human body image of the installation and maintenance personnel through a camera, and performing preprocessing;

[0063] S2, using object detection algorithms (such as YOLO, R-CNN, Transformer-based (such as DETR, Swin Transformer, etc.)), constructing an object detection model (such as convolutional neural network CNN, YOLO series, R-CNN, etc.), and inputting the preprocessed human body image into the object detection model for recognition and feature extraction;

[0064] S3. Input the advanced features into a classification decision tree model (the classification decision tree model is used to determine whether the dressed features obtained by using the object detection algorithm meet the dressing standards. It can be judged in the most primitive way based on program rules, such as using a rule engine and manually writing judgment rules, and at the same time, it can also be judged using a decision tree based on machine learning. Among the extracted features, there are both discrete values and continuous values. For example, whether a safety helmet is worn and whether the safety helmet is damaged are discrete values for judging the presence or absence of features; while the positions of these features in the image are continuous values. Therefore, CART analysis (Classification And Regression Trees) is needed to construct a decision tree) to determine whether the extracted features meet the specified dressing standards. If the extracted features meet the specified dressing standards, proceed to step S4; otherwise, output a warning signal to prohibit the operation task from continuing;

[0065] S4. Simulate a high-voltage environment through a preset test device to detect the warning functions of the safety helmet and the voltage detector;

[0066] S5. If all safety specifications and tool functions are detected to be qualified, the operation is allowed; if the detection result is unqualified, a warning signal is issued and the operator is prohibited from continuing the operation. Only when the detection is qualified can it be released. If the safety specifications are not met, the operation is forced to stop to ensure that the safety protection measures are implemented in place.

[0067] The present invention can improve the efficiency and accuracy of safety protection inspection before the operation of installation and maintenance personnel, reduce the errors and omissions of manual inspection, and ensure that the operation personnel strictly abide by the safety protection specifications before going out for work. Integrate safety detection with the daily operation process of installation and maintenance personnel, automatically check their safety protection equipment before the installation and maintenance personnel go out for work, strengthen the safety management link in the operation process, and ensure that the installation and maintenance personnel develop good safety protection habits. Incorporate intelligent technical means to conduct automated and normalized monitoring and analysis of the safety protection situation of installation and maintenance personnel before going out for work, timely discover and warn of potential safety hazards, and reduce the blind spots and loopholes of manual inspection.

[0068] In a preferred embodiment of the present invention, the method of using an object detection algorithm to construct an object detection model and inputting the preprocessed human body image into the object detection model for recognition and feature extraction is as follows:

[0069] Construct multiple object detection models according to requirements to extract features in the image. Each object detection model outputs the position, category information, and confidence of the corresponding features, obtaining multi-dimensional feature information. For example, 1. The categories detected by the safety helmet model are roughly: the heads of people wearing safety helmets, the heads of people not wearing hats, the heads of people wearing other hats, and the damaged parts of safety helmets; 2. The work uniform model: the work uniform itself, unzipped zippers, work uniform logos, and other features; 3. The safety belt model: the middle buckle of the safety belt, the shoulder strap, and the lower buckle.

[0070] The features extracted by the object detection model include the wearing situation of safety helmets, the damaged situation of safety helmets, the opening and closing situation of work uniform zippers, and the wearing situation of insulating gloves. On the basis of existing safety helmet and electric pen detection, the detection process and architecture of the present invention can be extended to more safety tools, such as high-altitude operation belts, insulating shoes, goggles, etc. By simply supplementing the corresponding sensors and recognition models, the system can be horizontally expanded.

[0071] In a preferred embodiment of the present invention, the previously obtained features are input into a classification decision tree model. The method for judging whether the extracted features meet the specified dressing standards is as follows:

[0072] The classification decision tree model is provided with a threshold configuration library, and the threshold configuration library stores the thresholds or judgment rules corresponding to each dressing specification;

[0073] Comprehensively judge different features (judgment is based on the classification decision tree, and the judgment rules and thresholds are obtained through training, not an explicit value. By creating samples of various different detection situations (taking images of actual detection situations), multi-dimensional detection features are output through the previously described model, and these features are labeled to form a data set to train the classification decision tree to make the model converge for compliance judgment).

[0074] In a preferred embodiment of the present invention, a preset test device is used to simulate a strong electricity environment to detect the warning functions of safety helmets and electric pens. The specific steps are as follows:

[0075] Set a certain intensity of electromagnetic field or near-electric environment, place the safety helmet and the electric pen in this environment, and the warning modules on the safety helmet and the electric pen emit warning sound signals and / or light signals;

[0076] Collect the warning signals emitted by the warning modules on the safety helmets and voltage detectors, and extract the characteristics of the warning signals (for sound warning detection, it is extracted through a microphone, and it is judged whether the sound conforms to the specific frequency range of the sound emitted by the corresponding voltage detector or safety helmet microphone. For the light signal, it is collected through a camera. First, the target detection algorithm is used to extract the position information of the safety helmet or voltage detector, and the corresponding part of the picture is intercepted. Then, the opencv tool is used to analyze whether there are light spots of the corresponding color and brightness values in the image to judge whether the warning light is on. Adding target detection can well exclude the interference of ambient light and greatly improve the detection accuracy);

[0077] Compare the extracted warning signal characteristics with the preset characteristic range. If the warning signal characteristics are not within the preset characteristic range, output the abnormal detection information of the safety helmet and voltage detector. Otherwise, output the normal detection information of the safety helmet and voltage detector.

[0078] In a preferred solution of the present invention, when the target confidence level is lower than the set value or the manual determination of the detection result is incorrect, the corresponding image or video segment is saved as a difficult example;

[0079] The developers make correct annotations for the collected difficult examples in the background and determine whether to include them in the training data;

[0080] After merging the difficult example library and the original data set, use the incremental training strategy to train and optimize the target detection model or decision tree determination model;

[0081] Replace the old version model with the updated model.

[0082] Collect and accumulate the data related to the safety protection before the operation of the installation and maintenance personnel, mine the data value, realize the digital management of the safety behavior of the installation and maintenance personnel, and provide data support for the subsequent optimization of safety decisions.

[0083] The combination of the difficult example automatic saving mechanism and the incremental training strategy enables the system to quickly absorb new scenarios and new cases, and continuously improve the detection accuracy and robustness. For different industries or different specification requirements, personalized customization can also be achieved by quickly configuring thresholds or adjusting training samples to meet diverse needs.

[0084] The present invention also provides a safety management and detection system for the personnel operation of the intelligent warehouse for material management based on the method described in the present invention, including a dressing code detection module and a safety tool test module. The dressing code detection module is used to detect whether the dressing of the installation and maintenance personnel is standard, and the safety tool test module is used to detect whether the functions of the tools carried by the installation and maintenance personnel are normal.

[0085] Strengthen the standardization, consistency and scientific nature of the safety management before the operation of the installation and maintenance team, improve the safety awareness of the installation and maintenance personnel through automated detection, and lay a good safety foundation for subsequent field operations.

[0086] In a preferred embodiment of the present invention, the dress code detection module includes a dress code feature extraction module and a dress code determination module connected in sequence.

[0087] The dress code feature extraction module includes a camera, an image preprocessing unit, a target detection model, and a feature data output interface connected in sequence. The camera is used to collect the human body image of the installation and maintenance personnel and transmit it to the image preprocessing unit. If there are extreme situations (night, strong light) in the environmental light or operation scene, an infrared camera and a multi-spectral camera module can be added to collect more dimensional image features and enhance the detection robustness of safety helmets and safety work clothes. In some high-temperature occasions, infrared imaging recognition can also be used to detect the human body or specific materials to assist in judging the wearing integrity of safety protection supplies.

[0088] If it is necessary to monitor the entire operation process in some high-risk scenarios, it can be extended to a real-time detection mode: cameras and sensors are arranged at the operation site for continuous monitoring (such as safety helmet detachment detection, danger warning), so as to realize the safety monitoring of the entire operation process.

[0089] The image preprocessing unit is used to perform quality optimization processing such as noise reduction, cropping, and distortion correction on the collected human body image. The target detection model extracts the features of the preprocessed human body image. Then, it is sent to the dress code determination module through the feature data output interface, and the feature information such as the bounding box and class confidence of the recognized targets such as the safety helmet and the position of the work clothes zipper is output to the dress code determination module.

[0090] The dress code determination module includes a classification decision tree model, a threshold configuration library, and a determination result output interface. The classification decision tree model is used to judge whether the extracted features meet the dress code standards (such as the safety helmet confidence threshold, the work clothes zipper confidence threshold, etc.) specified in the threshold configuration library, and outputs the judgment result through the determination result (qualified / unqualified) output interface.

[0091] In a preferred embodiment of the present invention, the dress code detection module further includes a dress code feature optimization & training module, which includes a hard example collection unit, a hard example screening unit, a model training unit, and a model update unit.

[0092] When the hard example collection unit detects that the target confidence is lower than the set value or the manual determination of the detection result is incorrect, it saves the corresponding image or video segment as a hard example. The hard example screening unit makes correct annotations (such as the specific wearing position of the safety helmet, judging whether it is damaged, etc.) on the collected hard examples by developers in the background, and determines whether to include them in the training data.

[0093] After the model training unit merges the hard example library with the original data set, it uses an incremental training strategy to train and optimize the object detection model or decision tree judgment model. The model update unit replaces the old version model with the updated model, achieving continuous optimization of the detection model and effectively enhancing the system's robustness to various complex dressing situations.

[0094] The hard example collection unit interacts closely with the judgment module to automatically obtain scenarios with insufficient confidence or manually determined to be misjudged. The hard example screening unit conducts manual confirmation and tagging in the background. The model training unit performs training after combining full-scale or incremental data, and replaces the new model into the system through the model update unit, forming a closed-loop optimization process.

[0095] In a preferred embodiment of the present invention, the safety tool test module includes a simulated high-voltage device, a microphone acquisition unit, a camera acquisition unit, a data processing unit, and an alarm judgment unit. The simulated high-voltage device is provided with a test point output interface. The microphone acquisition unit and the camera acquisition unit are used to collect alarm signals (such as sound signals, light signals, etc.) emitted by the alarm modules on the safety helmet and the voltage detector, and input them into the data processing unit.

[0096] The data processing unit is used to extract alarm signal features (including sound frequency, time-domain features, etc.) and input them into the alarm judgment unit. The alarm judgment unit compares the extracted alarm signal features with a preset feature range to judge whether the functions of the safety helmet and the voltage detector are normal. Based on the simulated high-voltage environment, it automatically detects the alarm capabilities of the safety helmet and the voltage detector, avoiding potential safety accidents caused by tool function failures.

[0097] The operator touches the voltage detector to the test point output interface; if the voltage detector is normal, it will emit a sound and / or light flash. The microphone collects the sound signal and the data processing makes a judgment; the camera collects the visual signal and the image recognition module judges whether there is a red light flash. If both or at least one of the alarms meet the expectations, it is judged that the voltage detector function is normal, otherwise it is judged as abnormal.

[0098] For example, bring the safety helmet close to the simulated high-voltage device; if the proximity electric sensor built into the safety helmet is triggered, an alarm sound will be generated; the microphone collects the audio signal, and the data processing unit judges whether it meets the expected alarm sound; if it meets, it is judged that the safety helmet function is qualified; otherwise, it is judged as unqualified and a warning is given.

[0099] Touch the voltage detector to the test point output interface; if the voltage detector is in a normal state, it will emit a sound alarm or a red light flash; the microphone and the camera collect alarm information simultaneously: Sound channel: Judge whether the audio features meet the alarm threshold; Visual channel: Judge whether the frame of the red light flash is detected;

[0100] If both alarms are detected or any alarm mode meets the expectation, it is determined that the function of the electric pen is qualified; otherwise, it is determined as unqualified.

[0101] For the alarm functions of electric pens or safety helmets, other detection methods such as current induction, Bluetooth sensing, and vibration detection can also be adopted, not limited to audio or visible light monitoring. In factory environments or remote warehouses, automated robotic arms or test benches can be designed to achieve batch and rapid detection of safety helmets or electric pens.

[0102] The computing power part of the system of the present invention is deployed in the intelligent warehouse for material management. If necessary, it can also be independently deployed in the departure area of the warehouse in the installation and maintenance site. All models can run locally in the computing unit. When it comes to model optimization and difficult case export, it is necessary to interact with the central server and enable cloud computing power.

[0103] The threshold configuration library of the decision tree can be flexibly configured according to the safety dressing specifications of different operators or enterprises; for detections that require more features (such as wearing shoe covers, goggles, etc.), the recognition range of the object detection model can also be expanded and corresponding thresholds can be set.

[0104] In the description of this specification, the descriptions with reference to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0105] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and purposes of the present invention. The scope of the present invention is defined by the claims and their equivalents.

Claims

1. A method for safety management and detection of personnel operations in a material management intelligent warehouse, characterized in that: The steps include: S1, register the identity information of the installation and maintenance personnel and the corresponding work tasks, collect the human body images of the installation and maintenance personnel, and perform preprocessing; S2, using the target detection algorithm to build a target detection model, and input the preprocessed human body image into the target detection model for recognition and feature extraction; S3, input the features in advance into the classification decision tree model to determine whether the extracted features meet the prescribed dress standards. If the extracted features meet the prescribed dress standards, proceed to step S4; otherwise, output a warning signal to prohibit the task from continuing; S4, simulates a strong electrical environment through a preset test device to test the alarm function of the safety helmet and the electrician's pen; S5: If all safety regulations and tool functions are tested and qualified, the operation is allowed; if the test results are unqualified, a warning signal is issued and the operator is prohibited from continuing the operation.

2. The method for safety management and detection of personnel operations in a material management intelligent warehouse according to claim 1, characterized in that: The method of using the target detection algorithm to build a target detection model and inputting the preprocessed human body image into the target detection model for recognition and feature extraction is as follows: Build multiple target detection models according to requirements to extract features from images. Each target detection model outputs the location, category information, and confidence of the corresponding feature to obtain multi-dimensional feature information. The features extracted by the target detection model include the wearing condition of the safety helmet, the damage condition of the safety helmet, the opening and closing condition of the zipper of the work clothes, and the wearing condition of the insulating gloves.

3. The method for safety management and detection of personnel operations in a material management intelligent warehouse according to claim 2, characterized in that: The method of inputting the advance features into the classification decision tree model and judging whether the extracted features meet the prescribed dress standards is as follows: A threshold configuration library is provided in the classification decision tree model, and the threshold configuration library stores the threshold or judgment rule corresponding to each dress code; Make comprehensive judgments on different characteristics.

4. The method for safety management and detection of personnel operations in a material management intelligent warehouse according to claim 1, characterized in that: Use the preset test device to simulate the strong electric environment and test the alarm function of the safety helmet and the electric test pen. The specific steps are as follows: An electromagnetic field or near-electric environment of a certain intensity is set, and a safety helmet and an electric test pen are placed in the environment, and the alarm modules on the safety helmet and the electric test pen emit an alarm sound signal and / or a light signal; Collect the alarm signals sent by the alarm modules on the safety helmet and the electrician's pencil, and extract the alarm signal features; The extracted alarm signal features are compared with the preset feature range. If the alarm signal features are not within the preset feature range, the abnormal detection information of the safety helmet and the electrician test pen is output, otherwise the normal detection information of the safety helmet and the electrician test pen is output.

5. The method for safety management and detection of personnel operations in a material management intelligent warehouse as claimed in claim 2 or 3, characterized in that: When the target confidence is lower than the set value or the manual judgment detection result is wrong, the corresponding image or video clip is saved as a difficult example; Developers will correctly label the collected difficult examples in the background and determine whether to include them in the training data. After merging the difficult example library with the original data set, the target detection model or decision tree judgment model is trained and optimized using the incremental training strategy; Replace the old version of the model with the updated one.

6. A material management intelligent warehouse personnel operation safety management and detection system based on the method of any one of claims 1-5, characterized in that: It includes a dress code detection module and a safety tool testing module. The dress code detection module is used to detect whether the dress of the installation and maintenance personnel is standard, and the safety tool testing module is used to detect whether the functions of the tools carried by the installation and maintenance personnel are normal.

7. The material management intelligent warehouse personnel operation safety management and detection system according to claim 6 is characterized in that: The dress code detection module includes a dress code feature extraction module and a dress code determination module connected in sequence; The dress code feature extraction module includes a camera, an image preprocessing unit, a target detection model and a feature data output interface connected in sequence, the camera is used to collect human body images of installation and maintenance personnel and transmit them to the image preprocessing unit, the image preprocessing unit is used to perform noise reduction, cropping and distortion correction on the collected human body images, the target detection model extracts the features of the preprocessed human body images, and then transmits them to the dress code determination module through the feature data output interface; The dress code determination module includes a classification decision tree model, a threshold configuration library and a determination result output interface. The classification decision tree model is used to determine whether the extracted features meet the dress standards specified in the threshold configuration library, and output the determination results through the determination result output interface.

8. The material management intelligent warehouse personnel operation safety management and detection system according to claim 7 is characterized in that: The dress code detection module further includes a dress code feature optimization & training module, wherein the dress code feature optimization & training module includes a difficult example collection unit, a difficult example screening unit, a model training unit, and a model updating unit; The difficult example collection unit saves the corresponding image or video clip as a difficult example when detecting that the target confidence is lower than a set value or the manual determination detection result is wrong; The difficult example screening unit correctly labels the collected difficult examples through developers in the background and determines whether to include them in the training data; After the model training unit merges the difficult example library with the original data set, it uses an incremental training strategy to train and optimize the target detection model or the decision tree determination model; The model updating unit replaces the old version model with the updated model.

9. The material management intelligent warehouse personnel operation safety management and detection system according to claim 6, characterized in that: The safety tool test module includes a simulated strong current device, a microphone acquisition unit, a camera acquisition unit, a data processing unit and an alarm judgment unit; The simulated strong current device is provided with a test point output interface, and the microphone acquisition unit and the camera acquisition unit are used to collect the alarm signals sent by the alarm modules on the safety helmet and the test pencil, and input them into the data processing unit; The data processing unit is used to extract the alarm signal characteristics and input them into the alarm judgment unit; The alarm judgment unit compares the extracted alarm signal characteristics with a preset characteristic range to judge whether the functions of the safety helmet and the electric tester are normal.

Images